DQC Logo
|

Private Cloud & On-Premises Deployments

The DQC Platform can be deployed in private cloud environments or fully on-premises—ensuring full data sovereignty and compliance with internal IT policies.


Deployment requirements

To run the DQC Platform in a private environment, the following infrastructure is required:

  • Kubernetes Cluster – Version ≥ 1.29.9

  • PostgreSQL Database – Version ≥ 14


Supported single sign-on (SSO) providers

The platform supports modern authentication methods:

  • Microsoft Entra (formerly Azure AD)

  • Okta

  • Any OpenID Connect-compliant provider


Bring your own LLM

You can connect your own LLM instances to the DQC Platform. The platform does not call a single fixed model: every request is routed to a capability class. One provider can cover all classes, and one model can serve several of them.

Supported providers

  • Anthropic Claude — via the Anthropic API or Amazon Bedrock

  • OpenAI — via the OpenAI API

  • Google Gemini — via Google Vertex AI

How many models do you need?

At minimum two models: STANDARD and FAST. Every other class routes to one of those two, so you can start with two and add dedicated models later where it pays off.

Model classes

Class

What it drives

Own model

Anthropic

OpenAI

STANDARD

Rule prediction, data enrichment, duplicate survivorship, coding assistant

Required

Claude Sonnet 4.6

GPT-5.6 Luna

FAST

Chat, coding assistant, LLM rules

Required

Claude Haiku 4.5

GPT-5.6 Luna

NANO

Rule descriptions, ruleset and mission naming

Optional — uses FAST

Claude Haiku 4.5

GPT-5.6 Luna

COLUMN_RELATION

Column matching when mapping system tables

Optional — uses FAST

Claude Haiku 4.5

GPT-5.6 Luna

PDF_PROCESSING

PDF extraction, schema chat, preview generation

Optional — uses STANDARD

Claude Sonnet 4.6

GPT-5.6 Luna

REASONING

Reserved for complex, multi-step analysis

Optional — uses STANDARD

Claude Opus 4.8

GPT-5.6 Terra or Sol

Minimum versions

  • Anthropic — Claude 4.x or newer. Sonnet 4.6 (STANDARD) and Haiku 4.5 (FAST) are the tested baseline.

  • OpenAI — GPT-5.4 or newer. GPT-5.4 is the floor and works across all classes. GPT-5.6 Luna is the better choice on cost and latency and is the recommended default.

  • Google — Gemini 3.x or newer. Gemini 3.5 Flash covers STANDARD and FAST, Gemini 3.1 Flash Lite covers NANO.

What the models must support

  • Tool calling — the platform runs agentic workflows, so every model must be able to call functions.

  • Structured outputs — results are validated against JSON schemas. On Gemini this requires version 3 or newer.

  • Binary PDF input — only for the PDF_PROCESSING class. The model must accept PDF files directly; Claude Sonnet 4.6 (or newer) is the tested option. Without it, PDF-based checks are unavailable, but the rest of the platform is unaffected.

Credentials for the chosen provider are configured during deployment — your onboarding contact will walk you through it.


undefined Notes

Private cloud & On-Premises Deployments | DQC